Say Goodbye to RAG! Build your own Streamlit App powered by Gemini 1.5 Pro's Multimodality 🚀

[@aricoleman5802]

Great video Keep up the good work

[@AviPatel68]

Thank you! Appreciate your comment and encouragement.

[@saeedzf]

Great video, please make more, thanks.

[@AviPatel68]

Thanks, will do!

[@victorkisala1279]

Can you do a video for a beginners

[@GustiAndika-e6w]

RAG still necessary to augmented user personalize data that not available public

[@ruidinis75]

That's a awsome job. I do have a doubt, when I load 3 documents and ask a question it tends to talk just about 1 document, the first or the last, do you know why ? They are different documents

[@AviPatel68]

Assuming you are using the non-VertexAI code then it's basically taking your 3 docs and stitching them together. if uploaded_files: text = "" for pdf in uploaded_files: pdf_reader = PdfReader(pdf) for page in pdf_reader.pages: text += page.extract_text() As good as Gemini 1.5 is, given how these docs are stitched it may provide response from 2 of the 3 reports. Try prompting like: what is said about the economy in each of the reports? break it down by each report. Providing the instruction to break it down by each report works for me, also perhaps try break it down by each of the 3 reports. Hope this helps.

[@insoo34]

Great video, thanks. Can I ask a question in error? Are path2 and path3 the same directory? I set path2 and path3 as my home directory and when I upload a pdf in path2 there is no error, but when I upload an image in path3 I get an error saying there is no image loaded. FileNotFoundError : [Errno 2] Why is this?

[@basese83]

me 2

[@ruidinis75]

You just need to create a directory and make it copy the image in there and then pass that directory and it works just fine

[@AviPatel68]

this was a work around i did to overcome the streamlit's file uploader. so, fpath is just the file name extracted when you upload the file/image. fpath2 is you providing the directory of the location of the images. fpath3 just puts together fpath2 and fpath. if the image i upload is bigcat.jpg, it's assigned to fpath. and if the directory where this image is located is /users/johndoe/images then fpath2 is equal to this directory. then, fpath3 puts these 2 things together so fpath3 would /users/johndoe/images/bigcat.jpg. i know it's confusing, if someone (which i know someone will) comes up with a more direct way to do it, please let us know. hope this helps.

[@yvessoke8558]

thanks

[@ruidinis75]

Can we have the type writer machine effect ? I hadd to create a function for that because the response is not a generator or stream type

[@AviPatel68]

Set stream=True when calling for the response. Something like: response = model.generate_content([prompt], generation_config, stream=True). Hope this helps.

[@paolosorgente7097]

Great videos, Google has never given me confidence like creating agents for evaluating documents, something that Claude does very well, months ago you made a video on this use case... now with Gemini could it be better than Claude what do you say?

[@AviPatel68]

I do believe Google DeepMind (Gemini) has the edge. Over the last many months, they have started to leverage some of their other own great work such as Mixture of Experts (MOE) architecture, etc. Personally, as long as Google has Demis Hassabis, they will continue to improve and do this the right way. My opinion.

[@Emily-k7r]

Isn't this one kind of RAG?